Computer Science 182a
Final Project

Final project proposal

• Overall project description
  In essence, this is what you'd like to do. Give some attention to how you (and I) might evaluate it. Also, you should consider the scope of your project here -- in particular, suggest ways that the project could be limited or simplified, if need be, as well as some ways in which it could naturally expand. It's completely OK for this to be a continuation of extensions you've pursued thus far -- or, it can be entirely new.



• System and Starting point
  Explain how you intend to start the project -- or, at least, the next stage of your project. Particulars of the start-up include the language (and perhaps platform/OS) under which you anticipate working and the source for input to your system (still images and where they'll come from, or, perhaps, video from somewhere). Any hardware requirements that I need to be aware of (so we can start setting things up)? Describe in as much detail as you can a starting piece of the project that you intend to begin with: this part will be due by April 10th. Also suggest a number of separate pieces that will comprise the whole system: what are your initial thoughts on the order in which to attack them? Are there any that will be more difficult than others?

Ideas for the final project

New! Scaffolded projects

If you'd rather pursue a well-scripted project than to create one from scratch, there are three ideas that I, personally, have felt were interesting and that have been used elsewhere. If you'd like, you could try one of these out (and that would probably add them to our regular projects in the future...):

• Object/image classification using the bag-of-features model
  
  This is one of the most effective current approaches to recognition. Below is a link to a scaffolded set of investigations that uses a standard data set (the "CalTech" images):
  bag-of-features classification using the CalTech dataset
  
  
• Stereo matching
  
  This is the classic reconstruction technique in which finding correspondences between two images allows for a 3d reconstruction of the scene. This also takes advantage of some standard datasets:
  The UNC Stereo-matching assignment
  
  
• Image-based localization
  
  This is a bit more open-ended: building a system that uses images to determine where you are within a previously-mapped environment. Below is a paper that describes an approach: the project would be to implement a subset of the paper sufficient to localize a laptop (or robot) that was in a familiar environment, e.g., the Galileo foyer or an HMC dorm room:
  Image Based View Localization System Retrieving from a Panorama Database by SURF

 Open-ended projects

More open-ended projects often fall under two broad headings: specific applications or comparative research. For the latter, the idea is to implement and investigate the relative performance of some set of vision algorithms. The key contribution is both summarizing the strengths and weaknesses of each and quantifying where they success and they they fail.

Algorithm implementation/testing/comparison

Svetlana Lazebnik has an excellent set of starting points for algorithmic investigations at this link. Broad topics include stereovision, geometric reconstruction, recognition, classification, detection, image segmentation, and video processing.

System-building ideas

As far as imagination allows, I suppose... Here are snapshots of three slides that listed a few of the ideas we have mentioned thus far:

Final project requirements

• Meeting
  

  Before April 15th, each team should meet with me to discuss their project's scope and ideas -- also, we will schedule an in-class presentation during one of the lecture days in late April.




• Presentation
  

  In one of the lecture sessions during the final two weeks of classes, each team or individual will present to the group a 10-15 minute overview of their project and approach and its context. For this presentation, you should present the progress made so far -- and what you hope to complete before the final demonstration.




• Write-up
  This is not due until May 7th -- the demonstration day is the more important (and earlier) deadline.
  
  The write-up is the formal documentation for the project. Although the format is not required to adhere to any particular form, the following sections outline a common approach to covering the important points:
  • An Introduction that motivates and describes the problem and the results at a high level.
  • A Background section that explains the basics of the algorithms you applied to the problem.
  • A System Description section that provides the details of how you constructed your system, how it works, and how you tailored the algorithms described in the previous section to the problem at hand.
  • A Results section that describes how well the system performs. A format that often works well here is to first explain your evaluation techniques, provide their results, and then explain those results and what they say about the problem and about your approach(es) to it.
  • A Conclusion, usually very brief, in which you can summarize your system and the results. In addition, this is a chance to be less scientific in your opinions about the project and a chance to put it in the larger context of larger, more general problems (such as the general vision problem or a broad subfield).
  The write-up should include a link to an archive of your source code and other supporting materials -- this will serve as a starting point for future students who may want to pick up where you left off... .



• Demo
  The demonstrations will be done during the final class period for CS 182a, which is Thursday, April 28 at 9:35am. This will be the "official" demo of your system; even if you choose an algorithm to study, implement, and/or extend, this date will be the time to show its results. Your demo should show off the portions of the system that went beyond whatever foundation you used as a starting point.